Catálogo de publicaciones - libros

In this paper, inverse fuzzy models for uncalibrated visual servoing, in 3D Workspace, are developed and validated in a six degrees of freedom robotic manipulator. This approach does not require calibrated kinematic and camera models, as needed in classical visual servoing to obtain the Jacobian. Fuzzy modeling is used to identify the inverse Jacobian in the robot workspace. Robot control is achieved by means of using the inverse fuzzy models directly as the controller. Experimental results obtained in a PUMA robot performing eye-to-hand visual servoing demonstrate the validity of the approach.

We propose a real-time 3D modeling system using multiple stereo cameras which is a method for displaying a target object from an arbitrary view. Each capturing PC segments the objects and estimates disparity fields, then they transmit the data to a 3D modeling server. A modeling server generates 3D models of the objects and renders a video at the designated point of view. A main contribution in this system is a shape-recovery algorithm to derive a 3D scene description from 2D images. We propose an efficient volume carving algorithm using silhouette and disparity at a time that is more robust than conventional algorithms. The algorithm partitions space into an octree and processes it hierarchically so that execution time and space are dramatically reduced. The generated free-view video provides realistic images of dynamically changing scenes in real-time.

This article introduces an innovative visual registration pro-cess suitable for textureless objects. Because our framework is industrial, the process is designed for metallic, complex free-form objects containing multiple bores. This technique is based on a new contour descriptor, invariant under affine transformation, which characterizes the neighborhood of a closed contour. The affine invariance is exploited in the learning stage to produce a lightweight model : for an automobile cylinder head, a learning view-sphere with twelve viewpoints is sufficient. Moreover, during the learning stage, this descriptor is combined to a 2D/3D pattern, concept likewise presented in this article. Once associated, the 2D/3D information wealth of this descriptor allows a pose estimation from a single match. This ability is exploited in the registration process to drastically reduce the complexity of the algorithm and increase efficiently its robustness to the difficult problem of repetitive patterns. Evaluations on a cylinder head, a car door and a binding beam confirm both the robustness and the precision (about 3 pixel of mean reprojection error on the full model reprojection area) of the process.

Gait recognition has become a popular new biometric in the last decade. Good recognition results have been achieved using different gait techniques on several databases. However, not much attention has been paid to get major questions: how good are biometrics data; how many subjects are needed to cover diversity of population (hypothetical or actual) in gait and how many samples per subject will give good representation of similarities and differences in the gait of the same subject. In this paper we try to answer these questions from the point of view of statistical analysis not only for gait recognition but for other biometrics as well. Though we do not think that we have a whole answer, we content this is the start of the answer.

The majority of automatic fingerprint matching systems depends on the comparison of the local ridge characteristics (bifurcation and termination), and a critical step in fingerprint matching is to extract minutiae from the input image. In this work we propose a novel ridge following algorithm based on a robust image enhancement filtering. Several experiments are carried out, showing the performances of the proposed approach.

Most fingerprint recognition techniques are based on minutiae matching and have been well studied. However, this technology still suffers from problems associated with the handling of poor quality impressions. One problem besetting fingerprint matching is distortion . Distortion changes both geometric position and orientation, and leads to difficulties in establishing a match among multiple impressions acquired from the same finger tip. In this paper, according to the particularity of fingerprint distortion, we propose a combined radial basis function (RBF) model, which separately builds rigid and nonrigid transformations, for attacking the distortion problem. Combined RBF model provides more accurate mapping function between a possible matched-pair. Experiments on real data demonstrate the efficacy of the proposed method in improving the compensation of fingerprint distortion.

In this paper, several configurations for a hybrid face/ear recognition system are investigated. The system is based on IFS (Iterated Function Systems) theory that are applied on both face and ear resulting in a bimodal architecture. One advantage is that the information used for the indexing and recognition task of face/ear can be made local, and this makes the method more robust to possible occlusions. The amount of information provided by each component of the face and the ear image has been assessed, first independently and then jointly. At last, results underline that the system significantly outperforms the existing approaches in the state of the art.

The problem of high dimensionality in face verification tasks has recently been simplified by the use of underlying spatial structures as proposed in the Two Dimensional Principal Component Analysis, the Two Dimensional Linear Discriminant Analysis and the Coupled Subspaces Analysis. Besides, the Small Sample Size problem that caused serious difficulties in traditional LDA has been overcome by the spatial approach 2DLDA. The application of these advances to facial verification techniques using different SVM schemes as classification algorithm is here shown. The experiments have been performed over a wide facial database (FRAV2D including 109 subjects), in which only one interest variable was changed in each experiment: illumination, pose, expression or occlusion. For training the SVMs, only two images per subject have been provided to fit in the small sample size problem.

We present a novel 3D facial feature location method based on the Spin Images registration technique. Three feature points are localized: the nose tip and the inner corners of the right and left eye. The points are found directly in the 3D mesh, allowing a previous normalization before the depth map calculation. This method is applied after a preprocess stage where the candidate points are selected measuring curvatures on the surface and applying clustering techniques. The system is tested on a 3D Face Database called FRAV3D with 105 people and a widely variety of acquisition conditions in order to test the method in a non-controlled environment. The success location rate is 99.5% in the case of the nose tip and 98% in the case of eyes, in frontal conditions. This rate is similar even if the conditions change allowing small rotations. Results in more extremely acquisition conditions are shown too. A complete study of the influence of the mesh resolution over the spin images quality and therefore over the face feature location rate is presented. The causes of the errors are discussed in detail.

Empirical studies concerning face recognition suggest that faces may be stored in memory by a few canonical representations. Models of visual perception are based on image representations in cortical area V1 and beyond, which contain many cell layers for feature extraction. Simple, complex and end-stopped cells provide input for line, edge and keypoint detection. Detected events provide a rich, multi-scale object representation, and this representation can be stored in memory in order to identify objects. In this paper, the above context is applied to face recognition. The multi-scale line/edge representation is explored in conjunction with keypoint-based saliency maps for Focus-of-Attention. Recognition rates of up to 96% were achieved by combining frontal and 3/4 views, and recognition was quite robust against partial occlusions.